World Leaders
Military Serviece
and Propensity for War

MAJ Dusty Turner

Okay, enough of that

Grant
Grant

Kennedy
Kennedy

Nixon
Nixon

Bush
Bush

Grant
Grant

Kennedy
Kennedy

Nixon
Nixon

Bush
Bush

Are leaders with military experience more likely to lead their countries in a militarized dispute?

Disclaimer

The research presented in this document does not represent the official views of the Army. The opinions and findings expressed herein are solely those of the authors or contributors and do not reflect Army policy or endorsement.

But first

Why are you at Baylor and why are you trying to answer this question…?

Army

  • Engineer Officer
    • Training: Fort Leonard Wood Missouri
    • Platoon Leader: Hawaii (Iraq)
    • Company Commander: White Sands Missile Range, NM (Afghanistan)
  • Assistant Professor / Instructor
    • United States Military Academy, West Point, NY
  • Operations Research Systems Analyst (ORSA)
    • Center for Army Analysis: Fort Belvoir, VA

Baylor

  • Dr (COL R) Rodney Sturdivant

  • Applied Logistic Regression

  • Associate Professor and Director of the Statistical Consulting Center

  • Dancing with the Waco Stars

Dr Peter Campbell

  • Associate Professor of Political Science at Baylor University
  • PhD in Political Science from the University of Notre Dame
  • Author of “Military Realism: The Logic and Limits of Force and Innovation in the US Army
  • Expert in:
    • Civil-military relations
    • Insurgency and counterinsurgency
    • The just war tradition
    • Cyber warfare

Are leaders with military
experience more likely to
lead their countries in a
militarized dispute?

The Leader Experience and Attribute Descriptions (LEAD) data set

  • Information assembled from
    • Previously assembled data sets
    • Academic books and articles
    • Obituaries
    • Genealogical databases
    • Military and state archives
    • U.S. Presidential Libraries
    • Photographs

What I assume about you…

Statistics

R

Real World Question

What I assume about you…

Statistics

\(\log\left(\frac{p}{1-p}\right) = \beta_0 + \beta_1X_1 + \beta_2X_2 + \ldots + \beta_kX_k\)

What I assume about you…

Statistics

R

Real World Question

What I assume about you…

R

glm(formula = y ~ x, family = binomial(link = "logit"))

What I assume about you…

Statistics

R

Real World Question

What I assume about you…

Real World Question

Grant Bush Nixon Kennedy

What I assume about you…

Real World Question

Grant Bush Nixon Kennedy

What I assume about you…

Statistics

R

Real World Question

Now, back to the point…

Explore

Model Building Techniques

  • Forward Selection
  • Backwards Selection
  • All Subsets
  • Purposeful Selection

Purposeful Selection

  • Univariable Analysis

  • Initial Multivariable Model

  • Compare Models

  • Preliminary Main Effects Model

  • Main Effects Model

  • Preliminary Final Model

  • Adequacy and Model Fit

Purposeful Selection



  • Univariable Analysis
  • Preliminary Final Model
  • Adequacy and Model Fit

Univariable Analysis

univariable <- function(factor) {

  formula <- str_c("cwinit", "~", factor)
  
  glm(formula, data = data, family = binomial(link = "logit")) |> 
  broom::tidy() |> 
  select(term, estimate, p.value) |> 
  filter(term != "(Intercept)")

}

Univariable Analysis

univariable <- function(factor) {

  formula <- str_c("cwinit", "~", factor)
  
  glm(formula, data = data, family = binomial(link = "logit")) |> 
  broom::tidy() |> 
  select(term, estimate, p.value) |> 
  filter(term != "(Intercept)")

}

Univariable Analysis

univariable <- function(factor) {

  formula <- str_c("cwinit", "~", factor)
  
  glm(formula, data = data, family = binomial(link = "logit")) |> 
  broom::tidy() |> 
  select(term, estimate, p.value) |> 
  filter(term != "(Intercept)")

}

Univariable Analysis

univariable <- function(factor) {

  formula <- str_c("cwinit", "~", factor)
  
  glm(formula, data = data, family = binomial(link = "logit")) |> 
  broom::tidy() |> 
  select(term, estimate, p.value) |> 
  filter(term != "(Intercept)")

}

Univariable Analysis

univariable <- function(factor) {

  formula <- str_c("cwinit", "~", factor)
  
  glm(formula, data = data, family = binomial(link = "logit")) |> 
  broom::tidy() |> 
  select(term, estimate, p.value) |> 
  filter(term != "(Intercept)")

}


univariable(factor = "age") |>
  gt::gt() |> 
  gt::fmt_number()
term estimate p.value
age 0.01 0.00

Univariable Analysis

future_map_dfr(
  .x = factors_of_interest,
  .f = ~univariable(factor = .x),
  .progress = T
)

Univariable Analysis

term estimate p.value odds
year −0.01 <0.01 0.99
warwin 0.89 <0.01 2.44
rebelloss 1.21 <0.01 3.34
combat 0.48 <0.01 1.62
rebel 0.33 <0.01 1.39
rebelwin 0.46 <0.01 1.58
careerpolitician −0.32 <0.01 0.73
religion 0.74 <0.01 2.10
age 0.01 <0.01 1.01
warloss 0.40 <0.01 1.49
marriedinpower −0.38 <0.01 0.69
business −0.44 <0.01 0.64
medicine −0.87 <0.01 0.42
militarycareer 0.24 <0.01 1.27
milnoncombat 0.30 <0.01 1.35
polity21 −0.01 <0.01 0.99
numGPs 0.08 <0.01 1.09
divorced −0.24 <0.01 0.79
teacher 0.19 <0.01 1.21
parstability 0.25 <0.01 1.28
totalspouses −0.06 <0.01 0.94
married −0.31 <0.01 0.74
activist 0.14 <0.01 1.15
illegit −0.36 <0.01 0.70
childtotal −0.01 0.03 0.99
gender 0.34 0.07 1.41
aristocratlandowner −0.14 0.07 0.87
law −0.07 0.19 0.93
police −0.26 0.26 0.77
orphanbinary −0.14 0.28 0.87
officetenure1000 −0.01 0.41 0.99
bluecollar −0.04 0.50 0.96
royalty −0.04 0.52 0.96
scienceeng 0.04 0.70 1.04
random <0.01 0.85 1.01
journalism <0.01 0.99 1.00
creative 0.00 1.00 1.00

Interaction Terms

Interaction combat milnoncombat militarycareer
000 – – –
101 ☑️ – ☑️
100 ☑️ – –
010 – ☑️ –
011 – ☑️ ☑️

Final Model

term log_odds odds_ratio p.value
(Intercept) −3.71 0.02 <0.01
log_yearssincemidinit −1.16 0.31 <0.01
illegit1 −0.51 0.60 0.02
medicine1 −0.47 0.63 0.06
royalty1 −0.39 0.67 <0.01
age 0.01 1.01 0.02
officetenure1000 0.03 1.03 0.04
milnoncombat_combat_militarycareer101 0.26 1.29 0.07
parstability1 0.29 1.33 0.11
milnoncombat_combat_militarycareer011 0.33 1.38 <0.01
yearssincemidinit 0.35 1.42 <0.01
milnoncombat_combat_militarycareer010 0.43 1.53 0.01
creative1 0.55 1.73 0.04
rebelloss1 0.70 2.00 <0.01
milnoncombat_combat_militarycareer100 0.87 2.40 <0.01

Final Model

term log_odds odds_ratio p.value
(Intercept) −3.71 0.02 <0.01
log_yearssincemidinit −1.16 0.31 <0.01
illegit1 −0.51 0.60 0.02
medicine1 −0.47 0.63 0.06
royalty1 −0.39 0.67 <0.01
age 0.01 1.01 0.02
officetenure1000 0.03 1.03 0.04
milnoncombat_combat_militarycareer101 0.26 1.29 0.07
parstability1 0.29 1.33 0.11
milnoncombat_combat_militarycareer011 0.33 1.38 <0.01
yearssincemidinit 0.35 1.42 <0.01
milnoncombat_combat_militarycareer010 0.43 1.53 0.01
creative1 0.55 1.73 0.04
rebelloss1 0.70 2.00 <0.01
milnoncombat_combat_militarycareer100 0.87 2.40 <0.01

Final Model

Interactions

Outliers


LogisticDx::dx(final_model, byCov=T)


\[\begin{align*} \Delta \hat{\beta} & : \text{The change in } \hat{\beta} \text{ when an observation is removed.} \\ \Delta \chi^2 & : \text{The change in the Pearson chi-square statistic with observation removed.} \\ \Delta D_i & : \text{The change in the deviance statistic when an observation is removed.} \\ \mathcal{h} & : \text{Leverage.} \\ \mathcal{sdr} & : \text{The standardized deviance residual.} \end{align*}\]

Outliers

So lets talk…

  • Data Issues
  • Causality
  • We only talked US Presidents

Thank you!